Accelerating Groundwater Flow Simulation in MODFLOW Using JASMIN‐Based Parallel Computing |
| |
Authors: | Tangpei Cheng Zeyao Mo Jingli Shao |
| |
Affiliation: | 1. Key Laboratory of Computational Physics, Institute of Applied Physics and Computational Mathematics, Beijing 100088, China.;2. High Performance Computing Center, Institute of Applied Physics and Computational Mathematics, Beijing 100094, China.;3. School of Water Resources and Environmental Science, China University of Geosciences (Beijing), Beijing 100083, China. |
| |
Abstract: | To accelerate the groundwater flow simulation process, this paper reports our work on developing an efficient parallel simulator through rebuilding the well‐known software MODFLOW on JASMIN (J Adaptive Structured Meshes applications Infrastructure). The rebuilding process is achieved by designing patch‐based data structure and parallel algorithms as well as adding slight modifications to the compute flow and subroutines in MODFLOW. Both the memory requirements and computing efforts are distributed among all processors; and to reduce communication cost, data transfers are batched and conveniently handled by adding ghost nodes to each patch. To further improve performance, constant‐head/inactive cells are tagged and neglected during the linear solving process and an efficient load balancing strategy is presented. The accuracy and efficiency are demonstrated through modeling three scenarios: The first application is a field flow problem located at Yanming Lake in China to help design reasonable quantity of groundwater exploitation. Desirable numerical accuracy and significant performance enhancement are obtained. Typically, the tagged program with load balancing strategy running on 40 cores is six times faster than the fastest MICCG‐based MODFLOW program. The second test is simulating flow in a highly heterogeneous aquifer. The AMG‐based JASMIN program running on 40 cores is nine times faster than the GMG‐based MODFLOW program. The third test is a simplified transient flow problem with the order of tens of millions of cells to examine the scalability. Compared to 32 cores, parallel efficiency of 77 and 68% are obtained on 512 and 1024 cores, respectively, which indicates impressive scalability. |
| |
Keywords: | |
|
|